File size: 3,681 Bytes
d323a32
 
3857e5a
 
 
 
d323a32
7f7be6e
3857e5a
bc314a3
79e15de
bc35806
c5c3a0e
79e15de
08da4b3
79e15de
b4dfacf
 
f93e55e
79e15de
cbba795
22df1f6
33c3dfd
 
5c918be
 
b4dfacf
5cbf752
33c3dfd
 
b4dfacf
2d55e63
f93e55e
553cb25
f93e55e
5a414f5
62ea27e
 
e5d628e
08da4b3
b4dfacf
 
f06d5d3
49d614e
 
f06d5d3
2f388b9
 
 
 
b4dfacf
 
407feb4
 
46a6aff
 
b4dfacf
 
 
 
 
 
76d4499
 
6b9d2d6
 
 
 
08da4b3
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
---
license: mit

tags:
- image-text-dataset
- synthetic-dataset
---


A collection of regularization / class instance datasets for the [Stable Diffusion v1-5](https://huggingface.co/runwayml/stable-diffusion-v1-5) model to use for DreamBooth prior preservation loss training. Files labeled with "mse vae" used the [stabilityai/sd-vae-ft-mse](https://huggingface.co/stabilityai/sd-vae-ft-mse) VAE. For ease of use, datasets are stored as zip files containing 512x512 PNG images. The number of images in each zip file is specified at the end of the filename.

There is currently a bug where HuggingFace is incorrectly reporting that the datasets are pickled. They are not picked, they are simple ZIP files containing the images.


Currently this repository contains the following datasets (datasets are named after the prompt they used):

Art Styles

* "**artwork style**": 4125 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**artwork style**": 4200 images generated using 50 DPM++ 2S a Karras steps and a CFG of 7, using the MSE VAE. A negative prompt of "text" was also used for this dataset.

* "**artwork style**": 2750 images generated using 50 DPM++ 2S a Karras steps and a CFG of 7, using the MSE VAE.

* "**illustration style**": 3050 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**erotic photography**": 2760 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**landscape photography**": 2500 images generated using 50 DPM++ 2S a Karras steps and a CFG of 7, using the MSE VAE. A negative prompt of "b&w, text" was also used for this dataset.

People

* "**person**": 2115 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**woman**": 4420 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**guy**": 4820 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**supermodel**": 4411 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**bikini model**": 4260 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**sexy athlete**": 5020 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**femme fatale**": 4725 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**sexy man**": 3505 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**sexy woman**": 3500 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

Animals

* "**kitty**": 5100 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**cat**": 2050 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

Vehicles

* "**fighter jet**": 1600 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**train**": 2669 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

* "**car**": 3150 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.

Themes

* "**cyberpunk**": 3040 images generated using 50 DDIM steps and a CFG of 7, using the MSE VAE.


I used the "Generate Forever" feature in [AUTOMATIC1111's WebUI](https://github.com/AUTOMATIC1111/stable-diffusion-webui) to create thousands of images for each dataset. Every image in a particular dataset uses the exact same settings, with only the seed value being different.

You can use my regularization / class image datasets with: https://github.com/ShivamShrirao/diffusers, https://github.com/JoePenna/Dreambooth-Stable-Diffusion, https://github.com/TheLastBen/fast-stable-diffusion, and any other DreamBooth projects that have support for prior preservation loss.